Indian Institute of Science Bangalore
etd@IISc Electronic Theses and Dissertations at Indian Institute of ScienceNot a member yet
6204 research outputs found
Sort by
Investigations on Generation of Multilevel 24-Sided Polygonal Voltage Space Vector Structures Without Vector Averaging for Variable Speed Drives
No full textInduction motors are mainly powered by two-level inverters in low-voltage, low-power drive applications. For medium and high voltage applications, a conventional two-level inverter needs high voltage rated switches, operates at high switching frequency to get better voltage quality, and produces high dv/dt at switching instants. While operating at six-step mode for full speed operation, a conventional two-level inverter produces low-order harmonics, mainly 5th, 7th, 11th, 13th, 17th, 19th, and so on. These lower-order harmonics produce torque pulsations, which can damage the motor and affects produced torque and power. Conventionally these low order harmonics are suppressed or eliminated by employing bulky and costly passive filters, which degrades dynamic performance of the motor. Another technique based on modified pulse width modulation is selective harmonic elimination, which suppresses fundamental voltage along with harmonics resulting in underutilization of the DC-link voltage. Multi-level inverters are widely employed in high power and high voltage motor drive applications due to lower harmonic distortion and lower dv/dt in the phase voltage. However, multi-level inverters produce hexagonal space vector structure (SVS) and introduce lower-order harmonics in phase voltage during operat= ion in overmodulation region. Also, as the levels increases, number of switches, number of capacitors, diodes and isolated power supplies also increases.
Polygonal SVS is a method for eliminating lower-order harmonics in full operating region. This thesis addresses the above-mentioned issues by generating a two-level and multi-level 24-sided polygonal SVS with real active vectors instead of switched average vectors. Each active vector is a real vector in contrast to switched average vectors in literature. The generation of real 24-sided vectors minimizes switching losses and improves the quality of phase voltage compared to switched averaged vectors technique. 24-sided polygonal SVS scheme eliminates lower order harmonics up to 19th order from motor phase voltage throughout the modulation range. The first work presents a method of generating 24-sided polygonal SVS comprised of 24 real active vectors and a zero vector. In the second work, a multilevel 24-sided polygonal SVS is presented, which suppresses higher order harmonics along with elimination of lower order harmonics from motor phase voltage. In the third work, an inverter circuit to generate a thirteen-level 24-sided polygonal SVS comprised of 288 real active vectors and a zero vector is presented. The SVS generated in third work is denser than the scheme pr= esented in the second work, which further improves output voltage quality, without altering the power circuit topology. In all above three works, vector timing computation is required, and reference vector is realized by time averaging nearest three vectors. To ensure the elimination of timing computation, a 24-sided polygon must be available for reference vector of any magnitude. In the fourth work, a variable speed induction motor drive to generate 24-stepped voltage waveform throughout modulation range is presented.
Inverter circuit is realized using primary and secondary inverters feeding an open-end winding induction motor. Primary and secondary inverters are implemented by cascading two-level inverters. DC sources for both inverters are realized using a simple star-delta transformer combination. The presented concepts are verified with laboratory prototypes. The presented work is suitable for medium voltage and medium power induction motor drive applications.PMR
Performance Characterization and Optimizations of Traditional ML Applications
No full textEven in the era of Deep Learning based methods, traditional machine learning methods with large data sets continue to attract significant attention. However, we find an apparent lack of a detailed performance characterization of these methods in the context of large training datasets. In this thesis, we study the systems behaviour of a number of traditional ML methods as implemented in popular free software libraries/modules to identify critical performance bottlenecks experienced by these applications. The performance characterization study reveals several interesting insights into the performance of these applications. We observe that the processor backend is the major bottleneck for our workloads, especially poor cache performance, coupled with a high fraction of CPU stall cycles due to memory latency. We also observed a very poor utilization of execution ports with only a single micro-op or no micro-op being executed for around 45% of the execution time. For the tree-based workloads, the CPU stalls due to badspeculation are also significant with values as high as 25% of CPU cycles. Then we evaluate the performance benefits of applying some well-known optimizations at the levels of caches and the main memory. More specifically, we test the usefulness of optimizations such as (i) software prefetching to improve cache performance and (ii) data layout and computation reordering optimizations to improve locality in DRAM accesses. These optimizations are implemented as modifications to the well-known scikit-learn library, which can be easily leveraged by application programmers. We evaluate the impact of the proposed optimizations using a combination of simulation and execution on a real system. The software prefetching optimization was implemented over ten workloads and it resulted in performance benefits varying from 5.2%- 27% on seven out of the ten ML applications while the data layout and computation reordering methods yielded around 8%- 23% performance improvement on seven out of eight neighbour and tree-based ML applications
Effect of Stimulus Normalization and Visual Attention at multiple scales of Neural Integration
No full textThe effect of visual attention on neural signals has been extensively studied using various techniques such as macaque neurophysiology and human electro/magneto encephalogram (EEG/MEG). Depending on the technique, different neural measures are typically used for studying attention. For example, in neurophysiology experiments involving macaques, many studies have focused on the modulation in spiking activity or the change in oscillatory power at different frequency bands such as alpha (8-12 Hz) or gamma (30-80 Hz) with attention, or the change in the relationship of spikes with these oscillations. In contrast, human EEG studies, in addition to studying alpha and gamma modulation, often use flickering stimuli that produce a specific neural response called steady-state visually evoked potential (SSVEP), which is also modulated by attention. However, due to the differences in stimuli and task paradigms in such studies, it is difficult to determine the effectiveness of these various neural measures for capturing attentional modulation. To address this, we designed a task paradigm which included both static and counterphase flickering stimuli to generate all the relevant neural measures (alpha/gamma power as well as SSVEPs) under identical recording conditions, which allowed us to compare their effectiveness in studying attention.
Since several reports suggest that attention modulates these neural measures through a canonical neural mechanism called normalization, in the first study of this thesis, we varied the normalization strength parametrically as a proxy for attentional modulation and tested its effect on various neural measures. We manipulated normalization strength by presenting static as well as flickering orthogonal superimposed gratings (plaids) at varying contrasts to two female monkeys while recording multiunit activity (MUA) and LFP from the primary visual cortex (area V1). We quantified the modulation in MUA, gamma (32-80 Hz), high-gamma (104-248 Hz) power, and SSVEP. Even under similar conditions, normalization strength was different for the four measures; and increased as: spikes, high-gamma, SSVEP, and gamma. However, these results could be explained using a normalization model, modified for population responses by varying the tuned normalization parameter and semi-saturation constant.
In the second part of the thesis, we tested the predictions of the gamma phase coding hypothesis in the context of stimulus contrast and visual attention. The gamma phase coding hypothesis posits that the intensity of the incoming stimulus is encoded in the position of the spike relative to the gamma rhythm. Using chronically implanted microelectrode arrays in the primary visual cortex of macaques engaged in an attention task while presenting stimuli of varying contrasts, we tested whether the phase of the gamma rhythm relative to spikes varied as a function of stimulus contrast and attentional state. We analyzed spikes and LFP from different electrodes and found a weak but significant effect of attention, but not stimulus contrast, on the gamma phase relative to spikes. Although we found a significant effect of attention, we argue that a small magnitude of phase shift as well as the dependence of phase angles on gamma power and center frequency, limits the potential role of gamma in phase coding in area V1.
In the third part of the thesis, we recorded EEG signals from 26 human participants while they were engaged in an attention task and analyzed alpha and gamma band powers for both static and flickering stimuli and SSVEP power for flickering stimuli. We report two main results. First, attentional modulation was comparable for SSVEP and alpha. Second, we found that non-foveal stimuli produced weak gamma despite various stimulus optimizations and therefore showed a negligible effect of attention although the same participants showed robust gamma activity for full-screen gratings. Thus, alpha and SSVEP won over gamma in capturing attentional modulation in human EEG. This result was in contrast to the findings of a comparable study in monkeys, where gamma and alpha won over SSVEPs. This study highlights the effectiveness of various neural measures in studying visual spatial attention and further implicates their usefulness in decoding behavior and attentional state in humans.DBT-Wellcome Trust India Alliance (Grant IA/S/18/2/504003), Tata Trusts, DBT-IISc Partnership Programm
Experimental Investigation of Electrons In and Above Liquid Helium
No full textElectrons on the surface of liquid helium form a nearly ideal 2-dimensional electron system (2DES). An electron density up to 2 × 10^9 cm-2, known as the critical electron density, can be achieved on the liquid helium surface, above which an electro-hydrodynamic (EHD) instability sets in, which results in the formation of MEBs. Due to this limitation in maximum possible density, only the classical liquid and solid phases of the 2DES can be accessed in this system.
But at the same time, on the surface of thin liquid helium film and with the multi-electron bubbles (MEBs), it may be possible to achieve high electron density than that of the critical electron density. This can allow the observation of quantum melting, i.e., the phase transition between the quantum solid to the liquid phase of the 2DES. Although extensive theoretical and experimental studies have already been done, the quantum melting transition has not been
achieved experimentally on these systems yet. In this thesis, we have used multiple new experimental approaches to obtain electron densities higher than what has been achieved before and to study the MEBs effectively.
First, we studied the temporal dynamics of the EHD instability and the effect of the applied electric field and charge density on the instability. The unstable wave vectors were determined experimentally, and their temporal growth was studied carefully. The determined unstable wave vectors were found to be a good match with the theoretically expected values obtained from the dispersion relation. At the same time, the analysis of the growth rate of unstable
vectors were found to be limited by the kinematic viscosity of the liquid helium.
Next, we investigated the lifetime of MEBs trapped on a dielectric surface and compared the result with previous results on free bubbles in bulk liquid helium. The reduced lifetime of trapped bubbles suggested an impact of convective heat flow around the bubbles on their lifetimes. Then we performed an experimental investigation that confirmed the effect of
convective heat flow direction inside the experimental cell on the lifetime of such trapped MEBs. Determination of the electronic phase inside an MEB is one of the biggest challenges of the time. Unfortunately, there is no direct way or technique for such investigation. We discussed how the MEB surface fluctuation with an external oscillating electric field could be observed, which may allow a possible way of studying the phase of the 2DES. We studied the surface fluctuations of electrically excited MEBs and observed different normal mechanical modes of the bubble wall. Then we extended our discussion on why liquid helium-4 is not a suitable medium to study the MEBs at low temperatures (below λ), where interesting phenomena occur, and how liquid helium-3, based on its physical property, can be a suitable replacement for this purpose. We generated MEBs inside liquid helium for the first time. The generated MEBs at 1.1 K were found to be stable with long lifetimes. This result opens the possibility of studying the MEBs at much lower temperatures where quantum properties dominate over classical for the 2DES.
Finally, we discussed the problem associated with achieving high electron density on the thin helium film and how integrating an NEA material as a substrate can help us overcome the problem. We fabricated NEA materials, i.e., cBN pellet, and optimized the rf sputtering deposition of cBN film. We performed a preliminary pick-up measurement on the charged thin helium with these materials as substrates, which showed some positive indications that need to
be confirmed with further advanced experimental investigations.INSPIRE, DST Indi
Variational Bayes Algorithms for mmWave and Massive MIMO-OFDM Systems with Low Resolution ADCs
No full textIn this thesis, we develop novel low-complexity algorithms for massive multiple-input multiple-output (MIMO) systems under practical non-idealities and theoretically analyze their performance. The first problem we consider is that of joint channel estimation and data decoding in uplink massive multiple-input-multiple-output (MIMO) systems with low-resolution analog-to-digital converters (ADCs) at the base station. The nonlinearities introduced by the ADCs make the problem challenging: in particular, the existing linear detectors perform poorly. Also, the channel coding used in commercial wireless systems necessitates soft symbol detection to obtain satisfactory performance. In this part of the thesis, we present a low-complexity variational Bayesian (VB) inference procedure to jointly solve the (possibly correlated) channel estimation and soft symbol decoding problem. We present the approach in progressively more complex scenarios, including the case where even the channel statistics are not available at the receiver. Then, we combine the VB procedure with a belief propagation (BP) based channel decoder, which further enhances the performance without any additional complexity. We numerically evaluate the bit error rate (BER) and the normalized mean squared error (NMSE) in the channel estimates obtained by our algorithm as a function of various system parameters, and benchmark the performance against genie-aided and state-of-the-art receivers. The results show that the VB procedure is a promising approach for developing low-complexity advanced receivers in low-resolution ADC based systems.
In the second problem, we consider the delay-domain sparse channel estimation and data decoding problems in a massive MIMO orthogonal frequency division multiplexing (MIMO-OFDM) wireless communication system with low-resolution ADCs. The high non-linear distortion due to coarse quantization leads to severe performance degradation in conventional OFDM receivers, which necessitates novel receiver techniques. Firstly, we derive the Bayesian Cramer-Rao lower bound (CRLB) on the mean squared error (MSE) in recovering jointly compressible vectors from quantized noisy underdetermined measurements. Secondly, we formulate the pilot-assisted channel estimation as a multiple measurement vector (MMV) sparse recovery problem, and develop a VB algorithm to infer the posterior distribution of the channel. We benchmark the MSE performance of our algorithm with that of the CRLB, and numerically show that the VB algorithm meets the CRLB. Thirdly, we present a soft symbol decoding algorithm that infers the posterior distributions of the data symbols given the quantized observations. We utilize the posterior statistics of the detected data symbols as virtual pilots, and develop an iterative soft symbol decoding and data-aided channel estimation procedure. Finally, we present a variant of the iterative algorithm that utilizes the output bit log-likelihood-ratios (LLRs) of the channel decoder to adapt the data prior to further improve the performance. We provide interesting insights into the impact of the various system parameters on the MSE and BER of the developed algorithms, and benchmark them against the state-of-the-art.
In the third problem, we present a novel model-and-data-driven channel estimation procedure in a millimeter-wave MIMO-OFDM wireless communication system. The transceivers employ a hybrid analog-digital architecture. We adapt techniques from a wide range of signal processing methods, such as compressed sensing and Bayesian inference, to learn the unknown sparsifying dictionary in the beamspace domain, as well as the delay-and-beamspace sparse channel. We train the model-based algorithm with a site-specific training dataset generated using a realistic ray tracing-based wireless channel simulation tool. We assess the performance of the developed channel estimation algorithm with the same site's test data. We benchmark the performance of our procedure in terms of NMSE error against an existing fast greedy method and two state-of-the-art algorithms, and empirically show that model-based approaches combined with data-driven customization outperform purely model based techniques by a large margin. This algorithm was selected as one of the top three solutions in the "ML5G-PHY Channel Estimation Global Challenge 2020" organized by the International Telecommunication Union.
In the last problem considered in this thesis, we study the problem of downlink (DL) sum rate maximization in codebook-based multiuser (MU) MIMO systems. The user equipments (UEs) estimate the DL channels using pilot symbols sent by the access point (AP) and feedback the estimates to the AP over a control channel. We present a closed form expression for the achievable sum rate of the MU-MIMO broadcast system with codebook constrained precoding based on the estimated channels, where multiple data streams are simultaneously transmitted to all users. Next, we present novel, computationally efficient, minorization-maximization (MM) based algorithms to determine the selection of beamforming vectors and power allocation to each beam that maximizes the achievable sum rate. Our solution involves multiple uses of MM in a nested fashion. Based on this approach, we present and contrast two algorithms, which we call the square-root-MM (SMM) and inverse-MM (IMM) algorithms. The algorithms are iterative and converge to a locally optimal beamforming vector selection and power allocation solution from any initialization. We evaluate the performance and complexity of the algorithms for various values of the system parameters, compare them with existing solutions, and provide further insights into how they can be used in system design
Deeply Scaled InAlN/GaN-on-Silicon High Electron Mobility Transistors for RF Applications
No full textWide bandgap gallium nitride (GaN) based high electron mobility transistors (HEMTs) are promising candidates for next-generation radio frequency (RF) power amplifier applications owing to high electron saturation velocity and large breakdown field. Conventional RF GaN HEMT stacks are epitaxially grown on semi-insulating silicon carbide (SiC) substrates. The high thermal conductivity of SiC and better crystal quality of epitaxial GaN-on-SiC helps in achieving excellent RF and power performance. However, the application of this technology in the emerging wireless communication sector is held back by the high cost of SiC substrates. Besides, GaN-on-SiC wafer size being limited to 6-inch diameter is a major bottleneck for large-scale production. Integrating GaN on silicon (Si) lowers the production cost and enhances scalability. This work aims to explore the RF and power performance of GaN-on-Si HEMTs. To enable deep scaling and achieve high cut-off frequencies, InAlN/GaN heterostructures are used as they offer high channel charge density with extremely thin InAlN barriers. This helps to maintain the aspect ratio while scaling down the device dimensions. One of the important parameters that limit the RF performance in deeply scaled devices is the Ohmic contact resistance. A dual approach for contact resistance minimization is attempted. First, using tetra-methyl ammonium hydroxide (TMAH) surface treatment and high-temperature annealing of a Ti-based metal stack, the Ohmic metal diffusion through the dislocations is enhanced to form TiN contacts in the GaN channel region. With this technique, a contact resistance of 0.23 Ohm-mm is achieved, which is three-fold lower compared to contacts fabricated without TMAH pre-treatment. Second, the polarization in the material is exploited to form Ohmic contacts using a novel Sc-based metal stack. A contact resistance of 0.39 Ohm-mm is obtained using this metal scheme. The mechanisms for the resistance reduction are investigated using transmission electron microscopy in each case, and a correlation is established between the microstructure and the contact resistance. Next, scaling studies are performed to enhance the cut-off frequency of the transistor. Using Ti-based Ohmic contacts for the source and the drain regions, a unity current/power gain cut-off frequency (fT/fmax) of 101/87 GHz is obtained in HEMTs with 200 nm gates. A record high effective electron velocity of 1.56 x 10^7 cm/s is estimated, which helps in achieving a state-of-the-art fT-LG product of 20.2 GHz-µm in GaN-on-Si devices. An fT/fmax of 240/47 GHz is achieved by scaling down the gate length to 40 nm. Finally, X-band (10 GHz) power performance is demonstrated in GaN-on-Si HEMTs. Two device configurations, the Schottky gate HEMT and the recessed gate metal-insulator semiconductor HEMT (MIS-HEMT), are explored. An output power density of 1.44 W/mm and a power added efficiency of 33% are obtained in the Schottky gate HEMTs with 100 nm gates. However, the low breakdown voltage limits the output power in these devices. Considerable breakdown enhancement is achieved in the MIS-HEMTs using an ex-situ metal organic chemical vapor deposited SiN gate dielectric. A high output power density of 2.75 W/mm and a power-added efficiency of 22% are demonstrated in the MIS-HEMT devices. To summarize, methods to minimize the Ohmic contact resistance in GaN-on-Si HEMTs are explored. Using InAlN/GaN heterostructures, RF and X-band power performance is demonstrated in highly scaled devices
A study of the evolution of the bulges and disks of spiral galaxies in interacting and isolated environments
No full textGalaxies are usually found in groups and clusters where they interact with each other gravitationally. These interactions affect the internal dynamics of the galaxies. In this thesis, we have studied the effect of flyby interactions and dark matter distributions on the evolution of bulges and disks of spiral galaxies. To understand the effect of flyby interactions on the bulges, disks, and spiral arms of Milky Way mass galaxies, we modelled disk galaxies with two types of bulges: classical bulges and boxy/peanut pseudo-bulges. We have performed N-body simulations of galaxy flybys of 10:1 and 5:1 mass ratios with varying pericenter distances. Using photometric and kinematic bulge-disk decompositions of the major galaxy at regular time steps, we found that the disks become shorter and thicker during flyby interactions. There is no effect of flyby interactions on the classical bulges. However, pseudo-bulges become dynamically hotter at the cost of hosting disks. We also found no effect of flyby interactions on the strength and the formation time of bar buckling. The tidally induced spiral arms are transient and are density waves in nature. Spiral arms form very soon after the pericenter passage of the galaxies and decay in two phases; the initial rapid winding phase and the subsequent slow winding phase. We confirmed that the spiral arms are the main drivers of the observed wave-like vertical breathing motion in the Milky Way, and the effect of tidal interactions does not directly induce breathing motion. In another project, using N-body simulations, we showed that the presence of oblate dark matter halos delay bar formation and so bar buckling is also delayed, but probate halos promote multiple buckling events. As a result of multiple buckling events, boxy/peanut structures in prolate halos show the maximum thickness. We have also studied the cosmic evolution of bulges since z=0.1 using SDSS data. We found that the disk-like pseudo-bulges are growing in number as the Universe is getting older. The pseudo-bulges appear optically diffuse compared to classical bulges and are commonly found in low mass galaxies. In the local volume, pseudo-bulges overcome the classical bulges even in bulge dominated galaxies, and so more than 75% of the local volume is rotation dominated. Finally, we have tested galaxy formation models of the cosmological simulation, Illustris TNG, using bulgeless galaxies. We selected Illustris TNG50 galaxies having mass greater than 10M and performed photometric decomposition to find bulgeless galaxies. We found that the bulgeless galaxies are metal-poor and have high specific angular momentum as compared to the galaxies with bulges and fall at the lower end of baryonic to dark matter mass ratio. Thus the TNG model is capable of producing a comparable fraction of bulgeless galaxies to those observed in the low redshift Universe
Physics-based Approach For Efficient & Reliable Enhancement-mode AlGaN/GaN High Electron Mobility Transistor (HEMT) Technology
No full textPower semiconductor devices have been the key to growth of power electronics market, and cover application areas ranging from mobile stations (commercial) to missile seekers (military). Continuously increasing demand for improving the power handling capacities and need for higher bandwidth capable devices has impelled the need for search of novel device structures and materials other than Silicon (Si). High electron mobility transistors (HEMT) based on Gallium Nitride (GaN) have emerged as one of the most promising candidates to replace Si based power semiconductor devices. AlGaN/GaN HEMTs offer several advantages over their Si counterpart, such as high electron mobility (~2000 cm^2/Vs), high sheet carrier density (~ 1E13 cm^-2), and higher critical electric field (~ 3MV/cm). These properties result in devices based on GaN to have superior and efficient ON-state performance, while achieving voltage handling capacities better than their Si counterparts. The theoretical material limits for GaN shows promising applications in power as well as RF domain. These application areas, combined with the widespread LED market, is expected to bring down the cost of development of GaN based devices to be competitive to that of Si-based devices. However, despite very high-power density and operational frequency figures already demonstrated for GaN HEMTs, there is yet to be wide-scale deployment of these devices. One of the major challenges in GaN HEMTs is the development of high performance yet reliable and fail-safe normally-OFF or enhancement-mode (e-mode) devices. Normally-OFF devices are the devices that conduct current only when a positive voltage is applied at one of its terminals, referred to as gate terminal. These devices have a positive threshold voltage (Vth) beyond which they conduct current or become ON. Such devices are desirable in power applications to improve reliability of the power electronic system. Further, extensive application of the GaN based devices is also limited by reliability challenges unique to this material system, such as dynamic ON resistance (Ron), ambient light dependent behavior, and hot electron induced degradation. This work follows a physics-based approach to demonstrate high performance and reliable e-mode devices using a device performance-reliability co-design approach. Physical insights gained into the mechanism governing e-mode operation, besides the reliability challenges, allowed the development of novel methods to demonstrate reliable normally-OFF AlGaN/GaN HEMTs in this work.
Firstly, to demonstrate high-performance e-mode AlGaN/GaN HEMTs, we have designed and developed a novel p-type and high-κ ternary oxide AlxTi1-xOy [1]-[2]. The p-type nature of the oxide resulted in an increase in Vth of the device, demonstrating a possibility to achieve positive Vth and hence e-mode operation. The oxide was developed in a thermal Atomic Layer Deposition system by depositing alternate films of Al2O3 and TiO2. By changing the deposition cycles of these films, the Al% and thickness of AlTiO could be precisely controlled. The p-type and high-κ nature of the resulting oxide was found to be a strong function of Al%. This allowed complete control over the p-type nature of developed AlTiO and hence threshold voltage of AlTiO gate oxide-based devices. Using the developed high-κ (25) and p-type Al0.5Ti0.5Oy as gate oxide, in conjunction with a thinner AlGaN barrier under gate, 600-V e-mode AlGaN/GaN HEMTs were demonstrated with performance metrices comparable to the best in literature. The HEMTs showed superior ON-state performance (ON current ~400 mA/mm and ON resistance = 8.9 ohm-mm) and gate control over channel (Ion / Ioff= 1E7, subthreshold slope = 73 mV/dec, and gate leakage < 200 nA/mm). Given that the developed p-type AlTiO was first of its kind, the next area of focus was to probe into the physical mechanism governing the 2-Dimensional Electron Gas (2-DEG) depletion or positive Vth shift achieved by the integration of these oxides in the gate stack. Given the wide band gap nature of AlTiO and AlGaN/GaN system, an electro-optical experiment-based method was used to probe the underlying mechanism [3], [4]. Experiments were carried out on devices with various gate oxides (Al2O3,TiO2, AlTiO) and GaN buffer stacks (varied carbon doping) with 365 nm UV exposure. These experiments revealed maximum negative Vth shift with UV exposure in AlTiO-gated devices. Moreover, the negative Vth shift was a function of Al% in AlTiO. Further, the negative Vth shift was established to be due to deionization of deep-level negative states in AlTiO, induced due to presence of Al at Ti sites ([Al]'Ti), at/near the oxide/nitride interface under the gate metal. The study thus identified the presence of negatively ionized deep-level states at room temperature to result in p-type doping of AlTiO, thereby leading to the positive Vth shift in AlTiO-based HEMTs.
Post demonstration of high-performance e-mode GaN HEMTs and the related physical mechanisms, the next part of the work dealt with gaining physical insights into reliability challenges plaguing AlGaN/GaN HEMTs with a view to achieving robust devices. Measure-Stress-Measure routines were followed to evaluate the dynamic ON resistance (dynamic Ron) of AlGaN/GaN HEMTs on carbon (C)-doped GaN buffer. The experiments revealed a unique stress time-dependent OFF-state drain-to-source critical stress voltage (Vcr), above which the dynamic Ron of AlGaN/GaN HEMTs increased significantly [5] – [7]. The Vcr was found to be a strong function of device design parameters, such as, gate-drain distance, field plate length, and passivation thickness. Moreover, the Vcr was observed for both Schottky and Metal-Insulator-Semiconductor (MIS)-HEMTs. With the help of detailed experiments with varying substrate bias, temperature and C-doping in GaN buffer, electron trapping in C-doping induced buffer acceptor traps is proposed and validated to be the source of the dynamic Ron degradation in these devices. This electron trapping is shown to be controlled by the electric field near gate connected field plate of the devices, as it modulates the trap ionization probability and injection of carriers into the GaN buffer. Thus, the HEMTs exhibit a Vcr beyond which high dynamic Ron is observed. Further experimentation on the gate bias dependence of the dynamic Ron of devices with different gate stacks revealed the carrier density in the GaN buffer to be a function of gate control over the channel [8]. This results in an OFF-state gate bias-dependent dynamic Ron degradation in AlGaN/GaN MISHEMTs. This is attributed to the MISHEMTs having an inferior channel control due to the insertion of the gate dielectric. Discovery of an electron trapping controlled dynamic Ron in GaN HEMTs encouraged us to examine the same under semi-ON state stressing as well [9]. Semi-ON state, where channel is in semi-ON condition, stresses the device in a condition where significant electron density exists in the presence of high electric field. This condition results in the generation of highly energized electrons, known as hot electrons. Dynamic Ron experimentations on GaN HEMTs under semi-ON conditions revealed that the interaction of hot electrons with the GaN buffer results in significant self-heating in the GaN buffer near the field plate edge and enhances electron de-trapping from these traps. On the other hand, trapping was found to be determined by field conditions near the field plate edge. This trapping-de-trapping process determines the net electron density trapped in the GaN buffer traps and thus determines the dynamic Ron of the HEMTs under semi-ON state stressing. Furthermore, the study revealed Schottky-HEMTs to have a better dynamic Ron behavior under semi-ON state stressing, as compared to the MISHEMTs. This was due to higher hot electron-induced self-heating and related de-trapping in the Schottky HEMTs. Besides analysing dynamic Ron behavior, reliability aspects related to device breakdown were also analysed [10]. Experiments revealed a slew rate dependent dynamic breakdown voltage in AlGaN/GaN HEMTs on a C-doped buffer. Detailed analysis revealed the role of electron transport through the C-doped GaN buffer in the observed HEMT breakdown behavior.
Further, besides gaining insights into the mechanisms governing dynamic Ron in GaN HEMTs, this work also demonstrates a methodology to improve the dynamic Ron of GaN HEMTs even in the presence of C-doping induced acceptor traps, which are introduced to improve the device breakdown. The developed methodology is based on the understanding gained from this work that electron trapping in the GaN buffer can be controlled by relaxing electric field magnitude near the field plate edge. The same was achieved in this work by incorporation of p-type Al0.5Ti0.5Oy as surface protection layer over SiNx passivation [7]. The proposed approach successfully relaxed electric field near field plate edge and resulted in mitigation of dynamic Ron in AlGaN/GaN HEMTs, even in the presence of C-doping induced buffer traps. This work thus resulted in the development of high performance and reliable AlGaN/GaN HEMTs on C-doped GaN-on-Si epi-stack, which was achieved through detailed physical insights into the governing mechanisms. Moreover, by solving the fundamental reliability challenge of dynamic Ron and demonstrating a novel AlGaN/GaN HEMT technology based on AlTiO, this work paves the way towards commercial deployment of a novel technology for high performance and reliable e-mode power HEMTs
Reliability Physics of Thin-Film Transistors
No full textThin-film transistor technology based on non-crystalline materials forms the workhorse
of large area electronics applications including display systems, sensor systems and novel
technologies including flexible electronics. A successful development and commercialisation
of any technology requires a thorough understanding of the physics and reliability
concerns revolving around that technology. Electrostatic discharge (ESD) is one of the
major reliability concerns in microelectronics industry and can plague the device development
at many stages. It is a high-field high-frequency phenomenon and involves charge
transfer from one body to another. Studying ESD behaviour also leverages studies on
non-equilibrium electro thermal behaviour of the device and highlights various high-field
effects taking place in the device under test. In this work, we study the ESD behaviour of
thin-film transistor technologies based on hydrogenated amorphous silicon (a-Si:H) and
organic materials.
In the first chapter, we give an overview of the current level of understanding vis-a-vis the
ESD behavior of non-crystalline materials based thin- lm technology platform. A brief
discussion on the ESD behavior and testing methodology of conventional silicon technology
is presented. We also discuss ESD issues in TFT technology, current roadblocks and
possible solutions. We also discuss other reliability issues that plague TFTs. Following
this, a detailed description, based on earlier studies, of the physical behavior, failure
characteristic and degradation behavior of hydrogenated amorphous silicon, metal oxide
based semiconductors, poly silicon and organics based semiconductors is presented. A
review of plethora of strategies that have been employed to enhance the ESD robustness
in these technologies, including novel designs and architectures is presented.
In the second chapter, we discuss, in detail, physical behavior of inverted staggered hydrogenated
amorphous silicon TFT technology under ESD stress conditions. Using electrical and optical techniques, device failure and TLP quasi-static I-V characteristics are discussed.
Raman spectroscopy is utilized to study any material variations with the stress
levels. Finally, we move on to discuss the impact of device design including impact of
device dimensions and architectural parameters including top passivation on the ESD
behavior of these devices.
As we move on to the third chapter, we study the device degradation of a-Si:H thin-film
transistors under the application of high field stress. I-V, C-V and raman spectroscopy
measurements are used to investigate the degradation mechanism. Threshold voltage
shift under moderate and high electric field in investigated and spatial variance of degradation
mechanism along the channel length is discussed. Variation of material properties
is studied. We also discuss the role of self-heating in device degradation and is studied by
varying the pulse width of stress pulses in nanoseconds range. We also discuss the performance
recovery mechanism through the application of thermal and gate bias anneal and
this has been investigated through a recursive cycle of stress- anneal and measurements.
Following these investigations, we report and study the phase transition behavior of
a-Si:H TFTs under high-field nano-second timescale electrical stress. This transition
is confirmed through a series of measurements including Raman Spectroscopy, Atomic
force microscopy, Scanning electron microscopy, Transmission electron microscopy and
I-V measurements. The observed behavior is attributed to avalanche generation, optical
phonon generation and localisation. We also study the case of drain underlap devices and
study how their behavior is different from conventional TFTs. Impact of pulse condition
including pulse width and channel dimensions on the onset of phase transition is also
explored. Interestingly, it is also found that the discussed phase transition mechanism
yields resultant nc-Si of quality at par with commercial methods.
At this point in the thesis, we have investigated the ESD and high-field reliability behavior
in a-Si:H TFTs. In the next chapter, we move our discussion towards incorporating
device architecture that improves the ESD robustness of a-Si:H TFT technology. The
discussed architecture is shown to improve the ESD robustness by 4-5 times with the
same area. We also investigate the physics of device behavior and explore the impact of
technological parameters on failure behavior. We also take a look at the pre breakdown
degradation behavior of this architecture. Finally, we take a look at the ESD behavior
of thin-film resistors.
In the next chapter, we discuss the ESD behavior of a-Si:H based diode-connected transistors. We discuss the ESD behavior as a function of stressing conditions, device dimensions
and on the application of negative ESD stress. We then discuss the instability behavior
of a-Si:H based diode-connected TFTs. This investigations assumes importance due to
the important role of these devices in switching and ESD protection circuits. Variations
in cut-in voltage of device under test is studied with application of the stress. DC I-V
measurements are used to explore the degradation behavior and shift in cut-in voltage as
a function of ambient temperature, pulse width and voltage levels is investigated. It is
also found that the degradation mechanism in these devices is different from the case of
conventional TFTs.
Till this point in this thesis, discussions have revolved around the behavior of a-Si:H
based devices. However, large area electronics based on novel classes of materials have
gained significant traction in recent years. One such class of materials is that of organic
semiconductors. Organic semiconductors also offer the advantage of cheaper fabrication
with lower thermal budget. This has enabled their application in flexible and printable
electronics. In the course of this work, we focus on pentacene as the model organic material.
In the seventh chapter, we discuss the ESD failure of pentacene channel OTFTs.
Charge transport mechanism at nano-second timescale is studied and orders of magnitude
difference are observed in DC and ESD timescales. Device failure is investigated through
SEM imaging and EDX spectroscopy. We also discuss the impact of self-heating behavior
along with the impact of channel dimensions and stressing conditions. It is observed that
the device failure is not due to semiconductor breakdown. We also study the impact of
introduction of self-assembled monolayer on charge transport and ESD failure. Finally,
we discuss the device failure for the case of high-voltage pentacene OTFTs with a 1μm
thick dielectric.
Following the investigation of ESD behavior of pentacene based OTFTs, we report high
frequency behavior of pentacene channel OTFTs through the exploration of transient
voltage waveforms in the eighth chapter. Pentacene OTFTs present a delayed response
to the applied signal due to the parasitic impedance involved. The device behavior is affected
by the bias stress effect and self-heating effect. It is also shown that as the channel
length increases, device responds faster. However, this behavior is shown to be present
at smaller channel length and as the channel length increases, the device impedance increases
and response gets slower
First thorough characterization of the structure, interactions and specificity of an archeal lectin with implication for TB infection
No full textDecades long studies on plant lectins carried out in this laboratory have contributed substantially to glycobiology and helped in the initiation and development of macromolecular crystallography in India. Subsequently the studies were extended to microbial lectins. It was realised that no lectin from archea has been thoroughly characterised. A genomic bioinformatic search led to the identification of several lectins from archea (Chapter 1).
An archeal lectin from Methanococcus voltae A3, christened Mevo lectin, has been cloned, expressed and purified. Crystallographic and solution studies of Mevo lectin and its complexes, the first effort of its kind on an archeal lectin, reveal a structure similar to β-prism I fold lectins from plant and animal sources, but with a quaternary association involving a ring structure with seven-fold symmetry. Each subunit in the heptamer carries one sugar binding site on the first Greek Key motif. The oligomeric interface is primarily made up of a parallel β-sheet involving a strand of Greek Key I of one subunit and Greek Key ΙΙΙ from a neighboring subunit. The crystal structures of the complexes of the lectin with mannose, αMan(1,2)αMan, αMan(1,3)αMan, a mannotriose and a mannopentose revealed a primary binding site similar to that found in other mannose specific β-prism I fold lectins. The complex with αMan(1,3)αMan provides an interesting case in which a few subunits have the reducing end at the primary binding site, while the majority have the nonreducing end at the primary binding site. The structures of complexes involving the trisaccharide and the pentasaccharide exhibit cross-linking among heptameric molecules. The observed arrangements may be relevant to the multivalency of the lectin. Phylogenetic analysis of amino acid sequences indicates that Mevo lectin is closer to β-prism I fold animal lectins than with those of plant origin. The results presented here reinforce the conclusion regarding the existence of lectins in all three domains of life. It would also appear that lectins evolved to the present form before the three domains diverged (Chapter 2).
Mannose-binding lectins can specifically recognize and bind complex glycan structures on pathogens and have potential as antiviral and antibacterial agents. Mevo lectin has specificity toward terminal α1,2 linked manno-oligosaccharides. Mycobacterium tuberculosis (M. tuberculosis) expresses mannosylated structures including lipoarabinomannan (ManLAM) on its surface and exploits C-type lectins to gain entry into the host cells. ManLAM structure has mannose capping with terminal αMan(1,2)αMan residues and is important for recognition by innate immune cells. Here, we aim to address the specificity of Mevo lectin toward high-mannose type glycans with terminal αMan(1,2)αMan residues and its effect on M. tuberculosis internalization by macrophages. Isothermal titration calorimetry (ITC) studies demonstrated that Mevo lectin shows preferential binding toward manno-oligosaccharides with terminal αMan(1,2)αMan structures and showed a strong affinity for ManLAM, whereas it binds weakly to Mycobacterium smegmatis lipoarabinomannan, which displays relatively fewer and shorter mannosyl caps. Crystal structure of Mevo lectin complexed with a Man7D1 revealed the multivalent cross-linking interaction, which explains avidity-based high-affinity for these ligands when compared to previously studied manno-oligosaccharides lacking the specific termini. Functional studies suggest that M. tuberculosis internalization by the macrophage was impaired by binding of Mevo lectin to ManLAM present on the surface of M. tuberculosis. Selectivity shown by Mevo lectin toward glycans with terminal αMan(1,2)αMan structures, and its ability to compromise the internalization of M. tuberculosis in vitro, underscore the potential utility of Mevo lectin as a research tool to study host-pathogen interactions (Chapter 3).
As mentioned earlier, Mevo lectin belongs to a highly conserved β-prism I fold lectin family and contains a single carbohydrate binding motif (132GXXXD136) on Greek Key I. Structural studies on complexes of mannose and mannose containing sugars with the lectin established that both Asp 134 and Asp 136 (part of conserved carbohydrate binding motif), are involved in the binding to carbohydrates. Asp 134 plays an important role in determining the specificity and affinity towards manno-oligosaccharides with αMan(1,2)αMan at the non-reducing end. To further elucidate the mechanism of the carbohydrate binding by Mevo lectin, two single mutants (D134A and D136A) and one double-mutant (D134/136A) were generated by site-directed mutagenesis. Analytical gel filtration results showed that all three mutants exhibited similar oligomeric state as the native lectin. X-ray crystallographic studies of Mevo lectin mutants revealed no major structural differences between the native lectin and the mutants. Binding analysis of the mutants by ITC indicated that Asp 136 is indispensable for the carbohydrate binding of Mevo lectin, whereas the D134A mutant retained its monosaccharide binding with reduced affinity compared to the native lectin. However, binding to manno-oligosaccharides having αMan(1,2)αMan at the non-reducing ends, such as mannotetrose, mannoheptose and ManLAM reduced very substantially. As expected, D134/136A double-mutant completely lost its carbohydrate-binding activity. These results suggest that Asp 136 is irreplaceable and Asp 134 plays an important role in determining the specificity and affinity of Mevo lectin to manno-oligosaccharides with terminal αMan(1,2)αMan residues and ManLAM (Chapter 4). It would appear that jacalin-like lectins or lectin domains with an aspartyl residue at the position corresponding to 134 are likely to interfere with uptake of M. tuberculosis by macrophages